The Clinical Trial That Asks Whether Your Current Treatment Should Exist

Medicine has spent decades building the machinery to prove that new interventions work. Nobody built equivalent machinery to prove when they should stop.

That oversight is now catching up with the field. A framework published in JAMA lays out the emerging discipline of deescalation, discontinuation, and deimplementation trials: studies designed not to add a therapy to the standard of care, but to interrogate whether the current standard of care deserves to stay there. The concept sounds modest. The regulatory and operational implications are anything but.

Consider what the evidence base actually looks like right now.

The Scale of What We Are Not Questioning

A peer-reviewed analysis published in JAMA Internal Medicine found that U.S. physicians estimated an interpolated median of 20.6% of overall medical care was unnecessary, including 22.0% of prescription medications and 24.9% of tests. The leading driver was not greed or ignorance. It was fear of malpractice. Physicians continued interventions they privately doubted because the system punishes doing less far more severely than doing more.

Benzodiazepines illustrate the problem precisely. In 2008, approximately 5.2% of U.S. adults between 18 and 80 used benzodiazepines, and among patients aged 65 to 80, nearly a third of that use was long-term. The clinical guidelines recommending short-term use only have existed for years. The prescribing pattern did not follow.

Spending on overtreatment alone, according to a definition framework maintained by the Center for Improving Value in Health Care, sits somewhere between $158 billion and $226 billion annually in the United States. That figure represents low-value care: services that offer little or no benefit while incurring costs and potential harm.

The standard clinical trial apparatus has no instrument to address this. Phase 3 trials prove that a drug works better than placebo or comparator. They do not prove that a currently approved drug, at its currently approved dose, for its currently approved duration, still makes sense when patients have been on it for three years beyond the studied window. Deescalation trials are built to answer exactly that question.

Why Trial Designers Have Avoided This Territory

The regulatory logic underpinning most clinical development assumes accumulation: earlier phase data supports later phase data, and the endpoint hierarchy reflects mounting evidence of benefit. Deescalation inverts the hierarchy. The trial question becomes whether less treatment produces outcomes that are noninferior, or in some cases actually superior, to continuing current treatment. That inversion creates genuine methodological friction.

Noninferiority margins in traditional trials are anchored to a historical placebo-controlled evidence base. In deescalation studies, the comparator is active ongoing treatment, and the question of what constitutes a clinically meaningful difference in the deescalation direction requires careful prespecification. Choose the margin too generously and you risk blessing a harmful reduction. Choose it too conservatively and you will never be able to demonstrate that stopping or scaling back is acceptable, regardless of how well patients do.

The FDA’s regulatory posture adds another layer of friction. Under 21 CFR 312.44, the agency retains authority to terminate an IND during Phase 1 if the trial poses unreasonable risk. That authority is appropriate. But it reflects a framework built around the concern that new interventions might harm patients. Deescalation trials carry a different risk profile: the intervention being tested is often the withdrawal or reduction of something patients and their physicians already believe is necessary. The perceived risk asymmetry runs in the opposite direction from conventional trials, which means sponsors face a steelman objection that regulators and IRBs tend to find persuasive regardless of the actual evidence.

The steelman is worth taking seriously: if you are wrong about deescalation, the patient loses a treatment that was working. That concern has some empirical weight in oncology, where premature treatment discontinuation has historically been associated with relapse. But the evidence base for that concern is far narrower than the territory over which the concern gets applied.

The Counterintuitive Case That Less Can Be Safer

Here is where the assumption collapses under the data.

A target trial emulation study published in CHEST Physician examined antibiotic de-escalation across 67 Michigan hospitals in 36,924 adults hospitalized with community-onset sepsis. These were patients receiving empiric broad-spectrum antibiotics with no evidence of multidrug-resistant organisms. The study found that de-escalation was safe, not merely noninferior in a statistical sense, but operationally safe across a population-level cohort. The fear that pulling back would destabilize patients did not materialize at scale.

That finding is not an isolated anomaly. The broader literature on deprescribing in older adults, shortened antibiotic regimens in pediatric infections, and de-intensification of antihypertensive therapy in patients who have achieved sustained control consistently shows that the harm from doing less tends to be smaller, and the harm from doing more tends to be larger, than intuition predicts. The field’s prior is wrong.

The principle at work here is what might be called the evidence inertia problem. Once an intervention enters clinical practice with regulatory approval or guideline endorsement, it acquires an evidentiary immunity that has nothing to do with ongoing benefit. Physicians continue it, payers reimburse it, and the clinical trial infrastructure has no funded mechanism to challenge it. Deescalation trials are the mechanism. The JAMA framework formalizes what the evidence has been suggesting for years: absence of ongoing benefit studies is not evidence of ongoing benefit.

Sponsors and CROs designing protocols in this space need to build three things their conventional trial templates do not include. First, the noninferiority margin needs to be justified not from a historical placebo effect estimate but from a patient-centered minimal important difference in the deescalation direction, which requires qualitative groundwork before the statistical analysis plan is written. Second, the safety monitoring framework needs to be calibrated to detect harm from under-treatment, not just over-treatment, meaning DSMBs need explicit stopping rules for both directions. Third, the regulatory strategy needs to engage the FDA early on labeling language: current prescribing information rarely specifies duration limits or de-escalation pathways, and a successful deescalation trial without a clear label update pathway generates academic credit without clinical impact.

The institutional resistance is real, but it is increasingly indefensible. When 22% of prescriptions are unnecessary by physicians’ own estimates, when benzodiazepine long-term use at age 65 runs at 31.4% despite decades of contrary guidance, and when the annual cost of low-value care clears $158 billion, the question of trial design philosophy stops being an academic exercise.

Every protocol that never asks whether the current dose is still justified is, implicitly, an endorsement of it. That endorsement is no longer scientifically neutral.

References

1. JAMA — “Deescalation, Discontinuation, and Deimplementation Trials”
2. JAMA Internal Medicine / PMC — “Physicians’ Views on Overtreatment: 20.6% of Overall Medical Care Unnecessary, Including 22.0% of Prescriptions”
3. JAMA Psychiatry — “Benzodiazepine Use in the United States: Long-term Use by Age Cohort”
4. Center for Improving Value in Health Care — “Low-Value Care Definition and Cost Estimates ($158B–$226B)”
5. eCFR — 21 CFR 312.44: FDA IND Termination Authority
6. CHEST Physician — “De-escalating Antibiotics Proves Safe for Patients with Community-Onset Sepsis” (67 Michigan hospitals, n=36,924)